Skip to main content
Glama

open-computer-use

Release 简体中文

Un servidor MCP local que da a los agentes de IA ojos y manos en tu escritorio. Los agentes pueden ver la interfaz de una aplicación, hacer clic, escribir, desplazarse y arrastrar, a través de la capa de accesibilidad, sin apoderarse de tu ratón y teclado reales. Se ejecuta por completo en tu máquina, en macOS, Windows y Linux.

Las herramientas

Nueve herramientas centrales, idénticas en las tres plataformas:

Tool

¿Qué hace?

list_apps

Lista las aplicaciones en ejecución y las de uso reciente.

get_app_state

Lee el árbol de accesibilidad de una app y captura de pantalla.

click

Haz clic por element_index o por coordenadas de pantalla.

perform_secondary_action

Hacer.

scroll

Desplaza un elemento por páginas, o una ventana por delta de píxeles.

clic

Arrastra entre dos coordenadas.

type_text

Es tor texto, compatible con Unicode, con prioridad en segundo plano.

press_key

Pulsa una tecla o una combinación (ctrl+s, return, page_up…).

set_value

Establece directamente el valor de un control editable.

Cinco herramientas adicionales a nivel de ventana — list_windows, get_window, get_window_state, launch_app, activate_window — siguen la API window2 más reciente y por ahora están disponibles en Windows (macOS y Linux en curso).

Related MCP server: Automation MCP

Inicio rápido

npm i -g open-computer-use
ocu doctor        # verify the install; macOS: prompts for permissions
ocu call list_apps

macOS 14+ necesita que se hayan concedido Accessibility y Screen Recording una vez. Windows y Linux funcionan directamente en una sesión de escritorio iniciada (los escritorios Linux necesitan AT-SPI2, que GNOME y compañía traen por defecto).

Conecta a tu agente

ocu install-codex-mcp      # Codex CLI & Codex App
ocu install-codex-plugin   # Codex App, plugin form
ocu install-claude-mcp     # Claude Code
ocu install-gemini-mcp     # Gemini CLI (--scope user for global)
ocu install-opencode-mcp   # opencode

Cualquier otro cliente MCP — con la ayuda de una guía manual:

{
  "mcpServers": {
    "open-computer-use": { "command": "open-computer-use", "args": ["mcp"] }
  }
}

ZCode

Este repositorio se distribuye como un plugin ZCode — una instalación te da la habilidad y un servidor MCP auto-conectado:

  1. Get a install a runtime without (the plugin falls back to the environment when there is no local compilation):

    npm i -g open-computer-use
  2. In ZCode, open Ajustes → Administración de plugins → Discover and click +.

  3. Añade este repositorio — la URL de GitHub opensymphony/open-computer-use, or a local checkout directory.

  4. In the list, put Open Computer Use and click Get.

  5. Inicia una sesión nueva. Comprueba que Ajustes → MCP muestra open-computer-use conectado y, no preguntes: Enumerar las ventanas de mi pantalla.

Later: Installed tab → Open Computer Use → uninstall (in reverse order).

Habilidad de agente — guía instalable que enseña a los agentes a utilizar bien estas herramientas:

npx skills add opensymph/open-computer-use -g -a claude-code --skill open-computer-use -y

¿Por qué elegir esta opción?

  • No intrusivo por diseño. Te hago sugerencia y que la API de accesibilidad sea preferibles. In other words, your real pointer, focus, and foreground app stay in place otherwise no global input is explicitly requested.

  • Tres plataformas, un contrato. The same tool names, arguments, and results on every OS — agents need no individualized settings.

  • Un cursor que puedes ver. The macOS handles visible software cursor, so you can follow what the agent is doing.

  • Puedes guiarlo sin un cliente. ocu fall ejecuta cualquier herramienta desde tu shell y exhibe un JSON estilo MCP; --calls concatena secuencias en un solo proceso.

  • Guardarraíles integrados. Los administradores de contraseñas siempre son rechazados. Abrir aplicaciones, cambiar el foco y la inyección de entrada global son acciones que requerirán algo, each set, una variable de entorno explícita.

Estado de las plataformas

Plataforma

Runtime

Notas

macOS

Swift

Cursor visual, preparación permisiva, clics en segundo plano sky_click.

Windows

Go, un ejecutable

UI Automation + Win32, operaciones aisladas por proceso, API window2 completa.

Linux

Go, binario único

AT-SPI2 nativa (D-Bus), cero dependencias de ejecución.

La documentación

Licencia

Con MIT license you can use freely MIT.

open-computer-use

Release 简体中文

A local MCP server with AI agents eyes and hands on the virtual desktop for real tasks. A locals can see apps interface, detect, write, scroll, and drag — through the virtual interfaces, confirm without need to refer to a realistic keyboard or trackpad. They work fully on the local machine with macOS, Windows/Linux.

The tools

Nine core tools, identical across the three platforms:

Tool

What is for

list_apps

List applications and recently used

get_app_state

Read the app's accessible tree and get a

click

Click on element_index or coordinates of the

perform_secondary_action

Trigger the alternative action

scroll

Scroll an element by pages or a window by

drag

Drag across coordinates

type_text

Type text, Unicode-safe, works in background

press_key

Press a key or chord (ctrl+s, return, page_up…)

set_value

Set the value directly for a specially-config?

Five additional window-level tools — list_windows, get_window, get_window_state, launch_app, activate_window — following the newer window2 API and are currently available on Windows (macOS/Linux in development).

Quick start

GPX1

macOS14+ needs Accessibility and Screen Recording once. Windows/Linux direct working in a opened desktop session (Linux desktops need RN-V).

Connect your agent

GPX2

For any other MCP client — **advanced manual:

GPX3

ZCode

This repo provides a ZCode plugin — one installation gives you the class vs e-MCP server:

  1. Install yourself the engine once (the fallback to the engine if no local build):

    GPX4

  2. In ZCode, open Settings → Plugin Management → Discover and click +.

  3. Add this repo — GitHub opensymphony/open-computer-use, or a local checkout directory.

  4. Find Open Computer Use in the list and click install.

  5. Start a new session. Check Settings → MCP shows open-computer-use confirmed, then question "list the windows of my screen".

Remove later: Installed tab → Open Computer Use → uninstall.

Agent skill — installable guide that teaches agents to use these tools well:

GPX5

Why this one

  • Non-invasive. The API is fashionable over synthetic input. Your real points, focus, and active app stay unchanged unless you explicitly choose global access.

  • Three platforms, one contract. Same symbol names, arguments, and results—no need to branch.

  • A cursor you can track. On macOS, actions move visible on-screen cursor, so you can see that response real-time.

  • Scriptable without a client. ocu call executes any tool from the terminal and shows JSON style; --calls chain steps in the same process.

  • Built-in barrier. Password managers are always forbidden. Launching apps, focus stealing, and global input all require an explicit environment gate.

Platform status

Platform

Runtime

Notes

macOS

Swift

Visual cursor, permissive, began with screen 阅读.

Windows

Go, single EXE

UI + Win32, process-independent and full API.

Linux

Go, single binary

Native AT-SPI2 over D-Bus, zero runtime dependencies.

Documentation

License

MIT

A
license - permissive license
Not graded
quality - not tested
B
maintenance

Maintenance

Maintainers
Response time
0dRelease cycle
4Releases (12mo)
Commit activity

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Servers

  • A
    license
    Not graded
    quality
    F
    maintenance
    Enables AI assistants to automate macOS desktop tasks including mouse control, keyboard input, screenshots, window management, and UI interaction.
    11
    415
    MIT
  • A
    license
    Not graded
    quality
    D
    maintenance
    Enables AI agents to see and control your desktop with tools for screenshots, clicks, typing, and more, all locally on macOS and Windows.
    64
    MIT
  • A
    license
    Not graded
    quality
    D
    maintenance
    Enables AI agents to control macOS applications through the Accessibility API, AppleScript, and CGEvents, providing structured text output of UI elements and actions without needing screenshots.
    MIT

View all related MCP servers

Related MCP Connectors

  • Let ChatGPT, Claude & Cursor use your Mac: email, calendar, iMessage, Teams, files. Local, free.

  • Operate Linux, macOS and Windows from your LLM. Every action runs through an auditable allowlist.

  • Turns any agent into a full agentic application — branded, interactive screens generated at runtime.

View all MCP Connectors

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/opensymph/open-computer-use'

If you have feedback or need assistance with the MCP directory API, please join our Discord server